Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!


Forgot your password?

Making Operating Systems Faster 667

mbrowling writes "In an article over at kernelthread.com Amit Singh discusses 'Ten Things Apple Did To Make Mac OS X Faster'. The theme seems to be that since you won't run into 'earth-shattering algorithmic breakthroughs' in every OS releases, what're you gonna do to bump your performance numbers higher? Although the example used is OS X, the article points out that Windows uses the same approach."
This discussion has been archived. No new comments can be posted.

Making Operating Systems Faster

Comments Filter:
  • by xplosiv ( 129880 ) on Thursday June 03, 2004 @10:37AM (#9325416)
    Check out www.blackviper.com [blackviper.com], it's one of the better sites dedicated to tuning and increasing performance of Windows 2000/XP
  • by wiggys ( 621350 ) on Thursday June 03, 2004 @10:42AM (#9325477)
    1) Don't install so much crap on your computer. 5 megapixel photos set as wallpaper along with Real Player, Gator Spyware Crap, Quicktime Task, HP scanner registration reminder sofware, webshots, Norton anything, MS office bar etc running on startup will make your nice shiny new computer run like an arthritic snail on sleeping tablets.

    2) Turn off some of the eyecandy. All those fades and whooshes and stuff don't actually do anything useful, they just consume CPU cycles and waste your time.

    3) Use Ad Aware and SpyBot regularly to keep scumware out of your computer. I had to clean up a PC this morning which had stopped working because the BASTARDS at NewDotNet wrote some software which fucked the TCP/IP stack backwards.

    4) Defrag regularly and run MSCONFIG to check what crap is sneaking back on to your Startup scripts.

    BTW, Windows 3.1 sitting on MSDOS 6.2 ran like shit of a stick on my old P133. I wonder if/how it would run on a modern system?
  • by pb ( 1020 ) on Thursday June 03, 2004 @10:44AM (#9325498)
    ... prelinking.

    What distro are you using?
  • by torpor ( 458 ) <ibisum@gm a i l . com> on Thursday June 03, 2004 @10:44AM (#9325503) Homepage Journal
    I dunno about all that but OS X doesn't seem slow at all to me.

    Try running LinuxPPC on your mac some day, and you will see a huge difference in general snappiness.

    I'm not saying OSX is un-usably slow, or even slow at all - heck my Rev. A tiBook, beaten and aged, is still all the computer I need, and I am very productive with it ... but I do have to admit that in all my computing experiences, OSX seems to be the one OS that is more 'acceptably mediocre', performance wise, than any other.

    On the register side of things, I can't for the life of me remember the full details, but I believe that the ABI for OSX only uses a sub-set of the PPC's full register set, and thus this means more swaps in/out ... that there are 'unutilized registers' in the PPC architecture when it is running OSX.

    This is separate from AltiVec, which is an instruction set, not just a register setup ...
  • Missing Step (Score:5, Informative)

    by baudilus ( 665036 ) on Thursday June 03, 2004 @10:44AM (#9325510)
    The number one thing they should do IMHO is reduce overhead. Using Microsoft Windows as an example, windows 98 has much less overhead than 2k, which in turn has much less than XP. A lot of it is eye-candy, which is all well and good, but those should be options that are OFF by default. XP differs from previous versions because it uses a 'shell' based gui (similar to KDE / GNOME, etc), which, while nice, is going to cause some system slowdown. Using the 'explorer' shell, which is heavily intergrated into the Windows OS, is the fastest, and should be the default. Then if people want to change it to look pretty [litestep.net] they can, by sacrificing speed (in slower machines).

    Stop adding services / features that are on by default, and you'll see a huge improvement in speed.
  • by Matthias Wiesmann ( 221411 ) on Thursday June 03, 2004 @10:57AM (#9325657) Homepage Journal
    Actually, if you had bothered with reading the article instead of repeating the old Apple has it easy with limited hardware cliché, you would have noticed that this is absolutely not related to driver performance.

    Only one optimisation presented is related to hardware drivers, and it is cache of what kernel extensions will probably be loaded. Most of the optimisations (basically lots of caching and dynamic defragmentation) could be implemented in Linux, regardless of the amount of supported hardware.

  • Re:That's 2 words. (Score:3, Informative)

    by Dibblah ( 645750 ) on Thursday June 03, 2004 @10:59AM (#9325685)
    You have an interesting definition of "justify". Besides, letting the VM do it's own thing with the buffer-cache does *much* better than stuffing RAM full of some random portion of disk that you think is 'important'.
  • Re:That's 2 words. (Score:4, Informative)

    by mbbac ( 568880 ) on Thursday June 03, 2004 @11:04AM (#9325729)
    Doesn't most unixes extensive use of cache really eliminate the benefits of that approach? I know Mac OS X will use almost all of however much physical RAM it's given.
  • Re:Faster? (Score:3, Informative)

    by baxissimo ( 135512 ) on Thursday June 03, 2004 @11:04AM (#9325734)
    By "refresh the desktop" do you mean that thing when all the icons disappear momentarily and then come back, possibly showing the generic icon for a moment before the actual icons appear?

    If that's what you mean by "refresh", then that's actually Windows Explorer (which the desktop is an instance of) crashing followed by a background process realizing it died and starting it back up.

    If that happens to you a lot then maybe you've installed some unstable shell extensions? Or maybe you're talking about something else. If it's specific to the XP theme I wouldn't know because I always revert to the "Classic" look first thing. The XP theme just looks like a cheap plastic toy.
  • by Miles ( 79172 ) on Thursday June 03, 2004 @11:09AM (#9325771) Homepage
    http://www.gentoo.org/doc/en/prelink-howto.xml [gentoo.org]

    A guide for gentoo, but the prelink program should be available for whatever distro you run.
  • by mbbac ( 568880 ) on Thursday June 03, 2004 @11:09AM (#9325772)
    5 megapixel photos set as wallpaper
    Actually, Quartz does an extremely good job of displaying 6.2 megapixel images on the desktop even on slow and old Macs.

    All those fades and whooshes and stuff don't actually do anything useful, they just consume CPU cycles and waste your time.
    Most of that is handed off to the GPU via Quartz Extreme.

    Defrag regularly
    HFS Plus already does that for me.
  • by tarunthegreat2 ( 761545 ) on Thursday June 03, 2004 @11:11AM (#9325802)
    I agree whole-heartedly. If Windows came installed 'Bare-Bones', there'd be a lot less annoyed people out there (but I'm sure we'd all miss Clippy)....however, that's one of the issues - who do you decide what should be an inherent part of the OS, and what shouldn't. Although you won't find anybody on slashdot propounding the beauty of having IE tied into Explorer, I know lots of AverageJoes who like the fact that they can just have that address bar on the TaskBar, and type a webaddress into it or a file path. Maybe "Where The Line Should Be Drawn" can be future Ask Slashdot article....
  • by Anonymous Coward on Thursday June 03, 2004 @11:13AM (#9325830)
    Are they really using it because that is 'what they had', or because they need a GOT (global object table) pointer to support dynamic loading? Shared objects and shared libs generally use a GOT to speed the relocation during the application load. Otherwise, you have to relocate ALL accesses, not just the unresolved symbols. Typically, overhead for GOT dedirection is closer to 3% or less....
    This last number comes from ELF documentation. Typically, ELF loaders (ant least in NIX) use a GOT.
  • by Anonymous Coward on Thursday June 03, 2004 @11:23AM (#9325993)
    Not everyone wants the equivalent of vi for every application in existence. There's a reason nobody (statistically speaking) uses Lynx to browse the Internet. It's ugly.

    That said I just wanted to point out that installed size is not always a good indicator of "bloat." My example would be MS Office vs. OpenOffice. The hard drive requirements for MS Office 2003 is significantly greater than that of OpenOffice 1.1. However, OpenOffice 1.1 requires nearly 3 times as much RAM than MS Office performing the same tasks (or even none at all, sitting idle OpenOffice is a massive hog.) To me RAM real estate is significantly more precious than hard drive.

    Now, you could use vi for your word processer and skip that "bloat" entirely, except that if that resume hit my desk it would meet the trashcan very quickly.
  • by phillymjs ( 234426 ) <slashdot AT stango DOT org> on Thursday June 03, 2004 @11:24AM (#9326003) Homepage Journal
    Has anyone tried using a RAMdisk as their OS drive?

    Many moons ago, it was possible to make a RAM disk on a Mac, install an OS on it, and (warm) boot from it. It would remain in memory and work perfectly as long as the computer wasn't shut down-- it could only be restarted. I tried it once or twice just to check it out, and the computer booted and ran like lightning compared to the normal hard drive boot.

    One of the utility suites back then (Central Point Utilities?) even had a feature where the machine would boot from a RAM disk with the utils on it, to fix the occasional really serious Mac problem.

    Booting from a RAM disk stopped being possible after Apple made a hardware change in newer Macs that had the side-effect of making the RAM non-persistent through warm-reboots (i.e., your RAM disk would go bye-bye). I forget exactly when it happened... perhaps after the first generation of Power Macs, when they went from using NuBus to using PCI?

    Here's another interesting fact. The Macintosh Classic, released in 1990, had System 6.0.8 (IIRC) burned into its ROM-- you could boot it disklessly from the OS in ROM by holding down Command-Option-O-X at startup. Nobody really knows what that feature was intended for.

  • by Doctor Crumb ( 737936 ) on Thursday June 03, 2004 @11:28AM (#9326073) Homepage
    I know it was a joke, but apple's GUI is rendered using the video card's processing power, not your CPU's. So such fancy effects are using cycles that would otherwise be idle, giving no performance hit at all, and making it look fricking cool at the same time.
  • more info (Score:3, Informative)

    by SilentT ( 742071 ) <.thetissilent. .at. .yahoo.com.> on Thursday June 03, 2004 @11:30AM (#9326087)
    Here's [vanemery.com] a "mini-HOWTO" that I found via google. I didn't read the whole thing, but it looks informative. Wikipedia's Ramdisk entry [wikipedia.org] had links to two stripped down knoppix distros that could be loaded into a ramdisk - Damn Small Linux [damnsmalllinux.org] (50 mb), and Feather Linux [berlios.de] (64 mb). I've never done anything with ramdisks (I'm a linux newbie, too) but they do sound pretty neat.
  • by squiggleslash ( 241428 ) on Thursday June 03, 2004 @11:31AM (#9326112) Homepage Journal
    OS X is a new development for PowerPC, not an improved 68K based system like OS 7
    This is incorrect. Mac OS X actually dates back to the mid-eighties and was originally developed for... yup, 680x0 based systems (I believe the original NeXT Cube had a 68020, can't be bothered to look it up now.) On top of which, in its original form, it was based on other pre-existing components such as the Mach kernel which date back even further.


  • Re:That's 2 words. (Score:3, Informative)

    by Moderation abuser ( 184013 ) on Thursday June 03, 2004 @11:33AM (#9326139)
    You're right, it does still have to be read once from the disk into RAM, and that's where I take my performance hit, at boot time but if you use dd with the right options on the partition rather than the filesystem, the disk will stream it at whatever the maximum streaming speed of the disk is, 50MB/s, 100Mb/s? It adds around 20secs to the boot time for me.

    Yup I have 2 copies of the apps in RAM. I have 4Gb of RAM, 2Gb of it as disk. My system doesn't swap, it still has 2Gb of RAM used as RAM and the performance is sensational.

  • Re:One word: (Score:5, Informative)

    by Graff ( 532189 ) on Thursday June 03, 2004 @11:36AM (#9326166)
    With buffer cache you're always going to have to touch the disk at some point and that point wastes billions of CPU cycles.

    Most systems nowadays use a DMA-type system (Direct Memory Access) [pcguide.com] which streams data directly from disk to memory without involving the CPU much at all. The real slowdown is not the CPU cycles getting wasted, it's that the CPU can't work on the particular data you need until it is loaded. During the DMA loading process your CPU could be using tons of cycles on other tasks that are not waiting on data.

    Smart read-ahead precaching and buffering attempts to ensure that your processes will not be data-starved. Yes, buffering can fall behind but overall it does considerably speed up a system.
  • by HeghmoH ( 13204 ) on Thursday June 03, 2004 @11:39AM (#9326207) Homepage Journal
    The only thing that's even remotely like what you're talking about is the Mach-O ABI and how it accesses global addresses.

    Mach-O the ABI (not to be confused with Mach-O the executable format, which is totally different) accesses global addresses via PC-relative addressing. This design decision was made back in the NeXT days, and made a lot of sense at the time. Unfortunately, the PowerPC doesn't have any support for PC-relative addressing, so the only way to do it is to use several instructions and induce a pipeline stall in the process. Depending on how a program is written, this problem can mean up to a 10% speed hit.

    That is the only brain-dead decision in the ABI that I'm aware of. It certainly makes good use of all registers, intelligently defines leaf procedures, and in general makes full use of the PPC architecture other than that one problem.

    Altivec includes both instructions and processors. That is one of the things that makes Altivec really cool, is that it has a shitload of vector registers that are totally separate from the other registers, and don't interfere in any way.
  • Re:Faster? (Score:4, Informative)

    by Anonvmous Coward ( 589068 ) on Thursday June 03, 2004 @11:44AM (#9326278)
    "If that's what you mean by "refresh", then that's actually Windows Explorer (which the desktop is an instance of) crashing followed by a background process realizing it died and starting it back up."

    Um, no. XP gives you an 'Explorer just crashed' message when it tanks. Heh my coworker next to me is actually having this 'explorer likes to crash regularly' problem. When you lose your taskbar and all your icons in the system tray disappear, then you know Explorer has gone south and restarted.

    Windows does have a 'refresh and rebuild the desktop' function. It's the same one they use to put your desktop icons back when you change video modes. (I.e. playing a game.) That's exactly what the person is describing.
  • by Areeves ( 598018 ) on Thursday June 03, 2004 @11:50AM (#9326350)
    Makes me all misty eyed over BeOS boot time of about 7 seconds to usability on a pIII 500. It CAN be done still.
  • by Anonymous Coward on Thursday June 03, 2004 @12:00PM (#9326516)
    No, their systems will run some competitor's software that didn't offer this option.
  • by boobert ( 7652 ) on Thursday June 03, 2004 @12:04PM (#9326560) Homepage
    My ibook can "sleep" for several days. Also I really like the fact that I can close it move around the house and my ssh sessions are still up when i open it up again.
  • by silicon not in the v ( 669585 ) on Thursday June 03, 2004 @12:04PM (#9326565) Journal
    Something I like about KDE (and Gnome may do this too--I've just never used it) in this area over Windows is their performance/eye candy slider. You don't have to go through finding which settings are the eye candy and turning them off one by one. KDE has a slider that you can drag toward features or performance, and it shows below that the settings that are being turned on or off.
  • by Have Blue ( 616 ) on Thursday June 03, 2004 @12:14PM (#9326697) Homepage
    It's not exactly "booting", but if you take a sleeping Apple notebook and wake it up, it'll usually be ready to use before you even get the lid all the way open. Re-establishing a wireless connection takes a few seconds, but local functionality all comes back on instantly.
  • by GooberToo ( 74388 ) on Thursday June 03, 2004 @12:30PM (#9326885)
    You're only telling half truths here. One of the speed ups that Microsoft does is, they analyze I/O and physically re-order the applications on disk to minimize I/O, thusly increasing startup speed.

    Furthermore, you assert that, "they are better engineered that the competition", which is completely false. It requires a superior engineered product to be crossplatform. In this case, IE has a speed advantage because it's NOT crossplatform, thusly allowing for more reasily available platform specific optimizations. It's also easy to forget that IE has less code, because it's far less compliant. Less code means less to load. Then, toss in the fact that significant portions of IE are cached by the OS during startup, gives IE a significant boon.

    Long story short, MS has many built in baises for starting up their applications which most applications are not able to benefit from.

    After that's all said and done, its the application performance, and not startup time that should really matter unless you're running a crappy OS. After all, these days, you should only need to start your applications once for the duration you're running your computer.

    If you really think you're comparing apples to apples, then I think we all understand why most Window's users benchmarks are ignored for what they are; invalid.
  • by argent ( 18001 ) <peter&slashdot,2006,taronga,com> on Thursday June 03, 2004 @12:46PM (#9327063) Homepage Journal
    And I turn off both sleep AND hibernate on my Windows laptop because the Windows sleep and hibernate don't work... the battery goes flat just as fast whether it thinks it's hibernating or not.

    The APM hibernation that the laptop's own BIOS implements works fine in FreeBSD, though. Wish Windows didn't take over that functionality.
  • by GooberToo ( 74388 ) on Thursday June 03, 2004 @12:50PM (#9327109)
    Well, all I know is that my own experience is different from yours. Not to mention, my experience is generally regarded as recreatable. That is, while I'm "logged in", my machine is technically worthless until it's finished starting up the system and doing all the things that Linux makes you do up front. In other words, when I have my desktop on Linux, I can immediately start using it. Under 2K and especailly XP, I have to wait, wait, wait before the system is responsive to my applicatin requests. That's the way MS designed their system and that's the way everyone experiences it. I guess this goes back to the perception versus reality difference. Like I said, it's up to you to decide if it's good or bad. I say, "indifferent". You seem to say, "good". Others say, "bad".

    It's worth noting, if nothing more than FYI points, there are ways to drastically speed up Linux's start up times. They range from using LinuxBios to changing out the init scripts for scripts which are are to run highly parallel. Last I heard, the init scripts alone, take off 10s of seconds. It's just that people would rather have UNIX and Linux compatibility.

    At any rate, I'm really not sure what you mean by, "USABILITY" being faster. If you mean the speed of the overall system as it relates to user responsiveness, then I suspect you have something wrong with your Linux configuration. Usabiity between the two systems should be equally high. Personally, my usability goes way down on Windows systems because it lacks so many of the powerful X features, out of the box anyways. But, I recognize that I'm not the typical win/linux user.

    Lastly, I must say that I find it interesting that you find XP to be faster than 2k. XP is widely regarded as being slower (yes, with everything turned off) than 2k, as far as the user interface is concerned.

    Some of these differences might center in how we're using our systems. My uses tend to be more of a workstation/desktop while you're may center completely around a MS-desktop solution.
  • by Overly Critical Guy ( 663429 ) on Thursday June 03, 2004 @12:53PM (#9327148)
    There are easier ways to enable these "features" than creating a ton of hoops for BOTH sides of users.

    What fucking hoops?

    Right-click My Computer->Properties->Advanced->Settings button.

    Choose either "Best Performance" or "Best Appearance." Or check each option individually. What a non-issue.

    If this was KDE, someone would have already answered with this, but because it's Windows, everyone just nods with the rest of the flock, "Baa, baa, yes, there are hoops to jump through, baa."

    Speaking of KDE, talk about fucking hoops. You've got a completely horrible control center, with three different areas for changing the looks of things like window styles, widget styles, and so on. Why the hell isn't that all integrated into one configuration dialog? Oh, I forgot, ease-of-use is a criticism we only reserve for non-issues on the Windows platform like checking a radio button to get rid of a blue theme.
  • Re:That's 2 words. (Score:1, Informative)

    by Anonymous Coward on Thursday June 03, 2004 @12:55PM (#9327159)
    Old-school Mac users may be familiar with this technique, except that it was once used for power conservation rather than performance. People would shoehorn a bare-bones version of System 7, a tiny word processor (Nisus was good for this) and a document or two into a 2 MB RAM disk, leaving just enough memory for the software to boot and run on a (typically 4 MB) PowerBook. The RAM disk would survive system sleep, restarts and even most system crashes. You could wake up the machine and get quite a lot of work done on the road without once spinning up the disk, and the battery life was phenomenal -- 2 or 3 times the normal duration, which was only a couple of hours at best, back in those days.

    Unfortunately, Mac OS X was a step backward in RAM disk technology on the Macintosh platform, and I don't think its possible to play this game anymore. Maybe power management has improved to the point where you wouldn't get much benefit out of it anyway. Anyway it was fun while it lasted.

  • by spells ( 203251 ) on Thursday June 03, 2004 @01:06PM (#9327249)
    You don't have to go through finding which settings are the eye candy and turning them off one by one
    On WinXP
    Right Click My Computer
    Choose Adjust for best performance OR
    Adjust for best appearance OR
  • Re:One word: (Score:3, Informative)

    by Smallpond ( 221300 ) on Thursday June 03, 2004 @01:32PM (#9327517) Homepage Journal
    Nope. Overall seek is the sum of average seek + rotational latency + single sector i/o (to account for head settling time). Seagate calls it average read time, although I like to use writes when measuring it. Track-to-track seek is not very interesting, unless you're a disk manufacturer, or a salesman playing games with specs.

    Brother. There's something I forget to mention. - Pi Patel

  • Re:Faster? (Score:3, Informative)

    by rhinoX ( 7448 ) on Thursday June 03, 2004 @01:45PM (#9327660)
    Wow, all this speculation. When windows refreshes like that it's because you changed a "system" setting, and set a "systemchanged" event. This causes applications that support it to refresh their settings from whatever store they have them in.

    This happens when you say, change your proxy settings (on or off, hit apply - bang, a refresh).

  • Re:FS Journaling (Score:3, Informative)

    by tgibbs ( 83782 ) on Thursday June 03, 2004 @01:55PM (#9327754)
    Filesystem journaling does not make the filesystem faster, and it's silly to suggest that it does. In fact, journaled filesystems are generally noticeably (one might say significantly) slower than non-journaled ones.

    As you'll see from this benchmark [macnn.com] Apple's implementation of journaling has generally negligible effect on performance, and some operations do in fact run faster.

  • Vague Steps (Score:2, Informative)

    by doorbot.com ( 184378 ) on Thursday June 03, 2004 @01:56PM (#9327768) Journal
    The number one thing they should do IMHO is reduce overhead.

    What exactly is "overhead"? It just sounds like a vague claim that the system is "inefficient" and needs to be "optimized".

    One advantage of XP over 2000 is that on XP you can disable the page file entirely, and Windows won't keep suggesting you enable it and/or complain because it's out of page file space (as happens if you set a 2 MB page file in 2000).

    My 1.6 Ghz/1 GB/80 GB laptop with XP with no pagefile is much more "responsive" than my
    1.6 Ghz/1 GB/80 GB desktop running 2000 with a pagefile. Windows seems to page memory to disk whether it's necessary or not; it will page out Thunderbird to disk, for example, just because it isn't the front-most application -- yet I have more than enough free RAM (let's say 256 in use, the rest used for buffers), the amount it saves by doing so is minimal at best. And I know you can set Thunderbird/Mozilla to stop Windows from paging it to disk, but should I have to do that for every app I want to use?
  • by wibs ( 696528 ) on Thursday June 03, 2004 @02:31PM (#9328156)
    Apple removed striping from everywhere in Panther. Quite a bit of it was replaced by brushed-metal. Even so, all it is doing is replacing one bitmap with another.

    Not true. The stripes are one graphic, yes. But metal windows are made up 9 graphics for the bevels, 1 more for the gradient, and another for the texture overlay. So a brushed metal window is actually rendering 10 more images than an Aqua window.
  • Re:That's 2 words. (Score:3, Informative)

    by Reziac ( 43301 ) on Thursday June 03, 2004 @02:50PM (#9328332) Homepage Journal
    That was a common trick back in the DOS era -- load whatever you could that you used all the time on a RAMdrive, then watch it BLINK onto the screen instead of crawling to the screen in slow motion.

    My first everyday computer (meaning it might be working all day long) was a 12MHz 286 with 1mb on the motherboard and 2mb on a RAMcard, which was set up as a RAMdrive (and had capacitors that let the data thereon survive a reset). The RAMdrive was where all the daily work happened, and it made a fantastic difference in performance.

    Another trick was using a RAMcard (which generally cannot be used as system RAM due to how the drivers worked) as Win16 swap space. In the era of slow HDs (then meaning somewhere around 1/100th of current disk speeds), that made a serious difference in performance.

  • by Anonymous Coward on Thursday June 03, 2004 @03:05PM (#9328510)
    As much as I love my Powerbook, I have to admit that this "instant-on" booting isn't exactly true. If you move between networks it takes several (5-10) seconds before you can enter your password to get to the desktop. If you leave network connections open (especially shares mounted through samba), it can take even longer.
  • by Eraser_ ( 101354 ) on Thursday June 03, 2004 @04:00PM (#9329034)
    Don't forget the ever harder Right Click desktop->Properties->appearance->Windows Classic. It almost doubles the amount of time it takes to get windows xp running twice as fast.

    The thing that drives me nuts is the constant harassment when you first install Windows XP for taking a tour and signing up for a .NET passport.
  • Re:Faster? (Score:2, Informative)

    by MCZapf ( 218870 ) on Thursday June 03, 2004 @04:09PM (#9329132)
    I think it's the apps themselves that refresh the taskbar system tray, not the other way around. That's why smarter apps, like ZoneAlarm, will put themselves back in the tray. But dumber apps - ones that only place themselves in the tray on startup - do not come back after Explorer crashes.
  • Sorry, that was COMPLETE misinformation.

    More digging indicates that these are a cache for the thumbnail images in the Desktop Pictures system preference. However the part about the desktop picture being stored on the GPU as a texture is still valid, as is the part about a 50MP image being no slower than a 50x50px GIF.

    I routinely set large (50MB, layers) photoshop files as my backdrop out of pure laziness and experience no slowdowns whatsoever as a result. (on a 1ghz 12" powerbook, 768mb RAM)

I THINK MAN INVENTED THE CAR by instinct. -- Jack Handley, The New Mexican, 1988.